Spaces:
Running
Running
Added Video feature.
#2
by
akaaku
- opened
app.py
CHANGED
@@ -4,25 +4,35 @@ import torch.nn.functional as F
|
|
4 |
from torchvision.transforms.functional import normalize
|
5 |
from huggingface_hub import hf_hub_download
|
6 |
import gradio as gr
|
7 |
-
|
|
|
8 |
from briarmbg import BriaRMBG
|
9 |
import PIL
|
10 |
from PIL import Image
|
11 |
from typing import Tuple
|
12 |
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
# model_path = "./model1.pth"
|
15 |
-
model_path = hf_hub_download("briaai/RMBG-1.4",
|
16 |
if torch.cuda.is_available():
|
17 |
net.load_state_dict(torch.load(model_path))
|
18 |
-
net=net.cuda()
|
|
|
19 |
else:
|
20 |
-
net.load_state_dict(torch.load(model_path,map_location="cpu"))
|
21 |
-
|
|
|
|
|
22 |
|
23 |
-
|
24 |
def resize_image(image):
|
25 |
-
image = image.convert(
|
26 |
model_input_size = (1024, 1024)
|
27 |
image = image.resize(model_input_size, Image.BILINEAR)
|
28 |
return image
|
@@ -32,28 +42,28 @@ def process(image):
|
|
32 |
|
33 |
# prepare input
|
34 |
orig_image = Image.fromarray(image)
|
35 |
-
w,h = orig_im_size = orig_image.size
|
36 |
image = resize_image(orig_image)
|
37 |
im_np = np.array(image)
|
38 |
-
im_tensor = torch.tensor(im_np, dtype=torch.float32).permute(2,0,1)
|
39 |
-
im_tensor = torch.unsqueeze(im_tensor,0)
|
40 |
-
im_tensor = torch.divide(im_tensor,255.0)
|
41 |
-
im_tensor = normalize(im_tensor,[0.5,0.5,0.5],[1.0,1.0,1.0])
|
42 |
if torch.cuda.is_available():
|
43 |
-
im_tensor=im_tensor.cuda()
|
44 |
|
45 |
-
#inference
|
46 |
-
result=net(im_tensor)
|
47 |
# post process
|
48 |
-
result = torch.squeeze(F.interpolate(result[0][0], size=(h,w), mode=
|
49 |
ma = torch.max(result)
|
50 |
mi = torch.min(result)
|
51 |
-
result = (result-mi)/(ma-mi)
|
52 |
# image to pil
|
53 |
-
im_array = (result*255).cpu().data.numpy().astype(np.uint8)
|
54 |
pil_im = Image.fromarray(np.squeeze(im_array))
|
55 |
# paste the mask on the original image
|
56 |
-
new_im = Image.new("RGBA", pil_im.size, (0,0,0,0))
|
57 |
new_im.paste(orig_image, mask=pil_im)
|
58 |
# new_orig_image = orig_image.convert('RGBA')
|
59 |
|
@@ -61,46 +71,164 @@ def process(image):
|
|
61 |
# return [new_orig_image, new_im]
|
62 |
|
63 |
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
# ''')
|
74 |
-
# with gr.Row():
|
75 |
-
# with gr.Column():
|
76 |
-
# input_image = gr.Image(sources=None, type="pil") # None for upload, ctrl+v and webcam
|
77 |
-
# # input_image = gr.Image(sources=None, type="numpy") # None for upload, ctrl+v and webcam
|
78 |
-
# run_button = gr.Button(value="Run")
|
79 |
-
|
80 |
-
# with gr.Column():
|
81 |
-
# result_gallery = gr.Gallery(label='Output', show_label=False, elem_id="gallery", columns=[1], height='auto')
|
82 |
-
# ips = [input_image]
|
83 |
-
# run_button.click(fn=process, inputs=ips, outputs=[result_gallery])
|
84 |
|
85 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
|
87 |
-
# block = gr.Blocks().queue()
|
88 |
|
89 |
gr.Markdown("## BRIA RMBG 1.4")
|
90 |
-
gr.HTML(
|
|
|
91 |
<p style="margin-bottom: 10px; font-size: 94%">
|
92 |
This is a demo for BRIA RMBG 1.4 that using
|
93 |
<a href="https://huggingface.co/briaai/RMBG-1.4" target="_blank">BRIA RMBG-1.4 image matting model</a> as backbone.
|
94 |
</p>
|
95 |
-
|
|
|
96 |
title = "Background Removal"
|
97 |
description = r"""Background removal model developed by <a href='https://BRIA.AI' target='_blank'><b>BRIA.AI</b></a>, trained on a carefully selected dataset and is available as an open-source model for non-commercial use.<br>
|
98 |
For test upload your image and wait. Read more at model card <a href='https://huggingface.co/briaai/RMBG-1.4' target='_blank'><b>briaai/RMBG-1.4</b></a>.<br>
|
99 |
"""
|
100 |
-
examples = [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
# output = ImageSlider(position=0.5,label='Image without background', type="pil", show_download_button=True)
|
102 |
# demo = gr.Interface(fn=process,inputs="image", outputs=output, examples=examples, title=title, description=description)
|
103 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
if __name__ == "__main__":
|
106 |
-
demo.launch(share=False)
|
|
|
4 |
from torchvision.transforms.functional import normalize
|
5 |
from huggingface_hub import hf_hub_download
|
6 |
import gradio as gr
|
7 |
+
|
8 |
+
# from gradio_imageslider import ImageSlider
|
9 |
from briarmbg import BriaRMBG
|
10 |
import PIL
|
11 |
from PIL import Image
|
12 |
from typing import Tuple
|
13 |
|
14 |
+
import cv2
|
15 |
+
import os
|
16 |
+
import shutil
|
17 |
+
import glob
|
18 |
+
from tqdm import tqdm
|
19 |
+
from ffmpy import FFmpeg
|
20 |
+
|
21 |
+
net = BriaRMBG()
|
22 |
# model_path = "./model1.pth"
|
23 |
+
model_path = hf_hub_download("briaai/RMBG-1.4", "model.pth")
|
24 |
if torch.cuda.is_available():
|
25 |
net.load_state_dict(torch.load(model_path))
|
26 |
+
net = net.cuda()
|
27 |
+
print("GPU is available")
|
28 |
else:
|
29 |
+
net.load_state_dict(torch.load(model_path, map_location="cpu"))
|
30 |
+
print("GPU is NOT available")
|
31 |
+
net.eval()
|
32 |
+
|
33 |
|
|
|
34 |
def resize_image(image):
|
35 |
+
image = image.convert("RGB")
|
36 |
model_input_size = (1024, 1024)
|
37 |
image = image.resize(model_input_size, Image.BILINEAR)
|
38 |
return image
|
|
|
42 |
|
43 |
# prepare input
|
44 |
orig_image = Image.fromarray(image)
|
45 |
+
w, h = orig_im_size = orig_image.size
|
46 |
image = resize_image(orig_image)
|
47 |
im_np = np.array(image)
|
48 |
+
im_tensor = torch.tensor(im_np, dtype=torch.float32).permute(2, 0, 1)
|
49 |
+
im_tensor = torch.unsqueeze(im_tensor, 0)
|
50 |
+
im_tensor = torch.divide(im_tensor, 255.0)
|
51 |
+
im_tensor = normalize(im_tensor, [0.5, 0.5, 0.5], [1.0, 1.0, 1.0])
|
52 |
if torch.cuda.is_available():
|
53 |
+
im_tensor = im_tensor.cuda()
|
54 |
|
55 |
+
# inference
|
56 |
+
result = net(im_tensor)
|
57 |
# post process
|
58 |
+
result = torch.squeeze(F.interpolate(result[0][0], size=(h, w), mode="bilinear"), 0)
|
59 |
ma = torch.max(result)
|
60 |
mi = torch.min(result)
|
61 |
+
result = (result - mi) / (ma - mi)
|
62 |
# image to pil
|
63 |
+
im_array = (result * 255).cpu().data.numpy().astype(np.uint8)
|
64 |
pil_im = Image.fromarray(np.squeeze(im_array))
|
65 |
# paste the mask on the original image
|
66 |
+
new_im = Image.new("RGBA", pil_im.size, (0, 0, 0, 0))
|
67 |
new_im.paste(orig_image, mask=pil_im)
|
68 |
# new_orig_image = orig_image.convert('RGBA')
|
69 |
|
|
|
71 |
# return [new_orig_image, new_im]
|
72 |
|
73 |
|
74 |
+
def process_video(video, key_color):
|
75 |
+
workspace = "./temp"
|
76 |
+
original_video_name_without_ext = os.path.splitext(os.path.basename(video))[0]
|
77 |
+
|
78 |
+
os.makedirs(workspace, exist_ok=True)
|
79 |
+
os.makedirs(f"{workspace}/frames", exist_ok=True)
|
80 |
+
os.makedirs(f"{workspace}/result", exist_ok=True)
|
81 |
+
os.makedirs("./video_result", exist_ok=True)
|
82 |
+
|
83 |
+
video_file = cv2.VideoCapture(video)
|
84 |
+
fps = video_file.get(cv2.CAP_PROP_FPS)
|
85 |
+
|
86 |
+
# まず、videoを読み込んで、./frames/にフレームを保存する
|
87 |
+
# fase, load video and save frames to ./frames/
|
88 |
+
def extract_frames():
|
89 |
+
success, frame = video_file.read()
|
90 |
+
frame_num = 0
|
91 |
+
with tqdm(
|
92 |
+
total=None,
|
93 |
+
desc="Extracting frames",
|
94 |
+
) as pbar:
|
95 |
+
while success:
|
96 |
+
file_name = f"{workspace}/frames/{frame_num:015d}.png"
|
97 |
+
cv2.imwrite(file_name, frame)
|
98 |
+
success, frame = video_file.read()
|
99 |
+
frame_num += 1
|
100 |
+
pbar.update(1)
|
101 |
+
video_file.release()
|
102 |
+
return
|
103 |
+
|
104 |
+
extract_frames()
|
105 |
+
|
106 |
+
# それぞれのフレームに対して処理を行う
|
107 |
+
# process each frame
|
108 |
+
def process_frame(frame_file):
|
109 |
+
image = cv2.imread(frame_file)
|
110 |
+
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
111 |
+
new_image = process(image)
|
112 |
+
# key_colorを背景にする
|
113 |
+
# set key_color as background
|
114 |
+
key_back_image = Image.new("RGBA", new_image.size, key_color)
|
115 |
+
new_image = Image.alpha_composite(key_back_image, new_image)
|
116 |
+
new_image.save(frame_file)
|
117 |
+
|
118 |
+
frame_files = sorted(glob.glob(f"{workspace}/frames/*.png"))
|
119 |
+
with tqdm(total=len(frame_files), desc="Processing frames") as pbar:
|
120 |
+
for file in frame_files:
|
121 |
+
process_frame(file)
|
122 |
+
pbar.update(1)
|
123 |
+
|
124 |
+
# frameからvideoを作成
|
125 |
+
# create video from frames
|
126 |
+
# first_frame = cv2.imread(frame_files[0])
|
127 |
+
# h, w, _ = first_frame.shape
|
128 |
+
# fourcc = cv2.VideoWriter_fourcc(*"avc1")
|
129 |
+
# new_video = cv2.VideoWriter(f"{workspace}/result/video.mp4", fourcc, fps, (w, h))
|
130 |
+
|
131 |
+
# for file in frame_files:
|
132 |
+
# image = cv2.imread(file)
|
133 |
+
# new_video.write(image)
|
134 |
+
# new_video.release()
|
135 |
+
|
136 |
+
# 上のコードをffmpyで書き直す
|
137 |
+
# rewrite the above code with ffmpy
|
138 |
+
ff = FFmpeg(
|
139 |
+
inputs={f"{workspace}/frames/%015d.png": f"-r {fps}"},
|
140 |
+
outputs={
|
141 |
+
f"{workspace}/result/video.mp4": f"-c:v libx264 -vf fps={fps},format=yuv420p -hide_banner -loglevel error -y"
|
142 |
+
},
|
143 |
+
)
|
144 |
+
ff.run()
|
145 |
+
# issue
|
146 |
+
# なぜかkey_colorの背景色が暗くなる
|
147 |
+
# idk why but key_color background color becomes dark
|
148 |
|
149 |
+
ff2 = FFmpeg(
|
150 |
+
inputs={f"{workspace}/result/video.mp4": None, f"{video}": None},
|
151 |
+
outputs={
|
152 |
+
f"./video_result/{original_video_name_without_ext}_BGremoved.mp4": "-c:v copy -c:a aac -strict experimental -map 0:v:0 -map 1:a:0 -shortest -hide_banner -loglevel error -y"
|
153 |
+
},
|
154 |
+
)
|
155 |
+
ff2.run()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
156 |
|
157 |
+
# 本当は透過の動画が良かったけど互換性がないのでボツ
|
158 |
+
# I wanted to make a transparent video, but it's not compatible, so I gave up
|
159 |
+
# subprocess.run(
|
160 |
+
# f'ffmpeg -framerate {fps} -i {workspace}/frames/%015d.png -auto-alt-ref 0 -c:v libvpx "./video_result/{original_video_name_without_ext}_BGremoved.webm" -hide_banner -loglevel error -y',
|
161 |
+
# shell=True,
|
162 |
+
# check=True,
|
163 |
+
# )
|
164 |
+
# クロマキー用なので音声いらないじゃん
|
165 |
+
# audio is not needed
|
166 |
+
|
167 |
+
# subprocess.run(
|
168 |
+
# f'ffmpeg -i "./video_result/{original_video_name_without_ext}_BGremoved.webm" -c:v libx264 -c:a aac -strict experimental -b:a 192k ./demo/demo.mp4 -hide_banner -loglevel error -y',
|
169 |
+
# shell=True,
|
170 |
+
# check=True,
|
171 |
+
# )
|
172 |
+
|
173 |
+
# ゴミ削除
|
174 |
+
# remove garbage
|
175 |
+
shutil.rmtree(workspace)
|
176 |
+
|
177 |
+
return f"./video_result/{original_video_name_without_ext}_BGremoved.mp4"
|
178 |
|
|
|
179 |
|
180 |
gr.Markdown("## BRIA RMBG 1.4")
|
181 |
+
gr.HTML(
|
182 |
+
"""
|
183 |
<p style="margin-bottom: 10px; font-size: 94%">
|
184 |
This is a demo for BRIA RMBG 1.4 that using
|
185 |
<a href="https://huggingface.co/briaai/RMBG-1.4" target="_blank">BRIA RMBG-1.4 image matting model</a> as backbone.
|
186 |
</p>
|
187 |
+
"""
|
188 |
+
)
|
189 |
title = "Background Removal"
|
190 |
description = r"""Background removal model developed by <a href='https://BRIA.AI' target='_blank'><b>BRIA.AI</b></a>, trained on a carefully selected dataset and is available as an open-source model for non-commercial use.<br>
|
191 |
For test upload your image and wait. Read more at model card <a href='https://huggingface.co/briaai/RMBG-1.4' target='_blank'><b>briaai/RMBG-1.4</b></a>.<br>
|
192 |
"""
|
193 |
+
examples = [
|
194 |
+
["./input.jpg"],
|
195 |
+
]
|
196 |
+
|
197 |
+
title2 = "Background Removal For Video"
|
198 |
+
description2 = r"""Background removal model developed by <a href='https://BRIA.AI' target='_blank'><b>BRIA.AI</b></a>, trained on a carefully selected dataset and is available as an open-source model for non-commercial use.<br>
|
199 |
+
For test upload your image and wait. Read more at model card <a href='https://huggingface.co/briaai/RMBG-1.4' target='_blank'><b>briaai/RMBG-1.4</b></a>.<br>
|
200 |
+
Also, you can remove the background from the video.<br>You may have to wait a little longer for the video to process as each frame in video will be processed, so using strong GPU locally is recommended.<br>
|
201 |
+
"""
|
202 |
+
|
203 |
# output = ImageSlider(position=0.5,label='Image without background', type="pil", show_download_button=True)
|
204 |
# demo = gr.Interface(fn=process,inputs="image", outputs=output, examples=examples, title=title, description=description)
|
205 |
+
demo1 = gr.Interface(
|
206 |
+
fn=process,
|
207 |
+
inputs="image",
|
208 |
+
outputs="image",
|
209 |
+
title=title,
|
210 |
+
description=description,
|
211 |
+
examples=examples,
|
212 |
+
api_name="demo1",
|
213 |
+
)
|
214 |
+
|
215 |
+
|
216 |
+
demo2 = gr.Interface(
|
217 |
+
fn=process_video,
|
218 |
+
inputs=[
|
219 |
+
gr.Video(label="Video"),
|
220 |
+
gr.ColorPicker(label="Key Color(Background color)"),
|
221 |
+
],
|
222 |
+
outputs="video",
|
223 |
+
title=title2,
|
224 |
+
description=description2,
|
225 |
+
api_name="demo2",
|
226 |
+
)
|
227 |
+
|
228 |
+
demo = gr.TabbedInterface(
|
229 |
+
interface_list=[demo1, demo2],
|
230 |
+
tab_names=["Image", "Video"],
|
231 |
+
)
|
232 |
|
233 |
if __name__ == "__main__":
|
234 |
+
demo.launch(share=False)
|